18:28
2026-07-31
runtimewire.com
large-language-models
Huawei releases 505B-parameter openPangu model trained on Ascend chips
Huawei released openPangu 2.0 Pro, a Mixture-of-Experts language model with approximately 505 billion total parameters and 18 billion activated per token, trained entirely on its Ascend AI processors.…